People are buying Mac Minis to run a lobster.

This is the first time I have felt like I’m living in the future since ChatGPT’s launch.

— Dave Morin, early Facebook PM

Not a new Apple product. Not a new ChatGPT feature. An open-source AI assistant built by a retired iOS developer, with a lobster mascot, that connects to your existing messaging apps and actually does things.

The Stories That Sell It

Before explaining what Clawdbot is, here’s what people are doing with it:

  • $4,200 car discount: AJ Stuyvenberg had his bot orchestrate a bidding war between dealerships. The bot negotiated via text while he watched. Full breakdown here.

  • Insurance fight: Someone’s bot accidentally started a dispute with their insurance company, leading to case reinvestigation

  • 180 million tokens: Federico Viticci (MacStories) burned through 180M tokens in roughly a week of experimenting

  • Nokia development: A user claimed they’re “literally building a whole website on a Nokia 3310” by texting their bot instructions

  • Mac Mini phenomenon: Multiple people bought Mac Minis specifically to run it 24/7

This feels like cheating. I texted my bot to book a restaurant while walking to the train. It checked availability, made the reservation, and confirmed with my wife. I just walked.

— @clawdbot community member

What It Actually Is

Clawdbot is an open-source personal AI assistant built by Peter Steinberger (@steipete). If you’re in iOS development, you know the name: he founded PSPDFKit in 2011, bootstrapped it to exit in 2021, and is legitimately an iOS legend.

The core idea: runs locally on your machine (Mac, Windows, Linux), connects to messaging apps you already use (WhatsApp, Telegram, iMessage, Discord, Signal, Slack), and actually executes tasks. 50+ integrations. Voice wake. Browser control. Shell access.

The lobster way

The mascot is a lobster. The community calls their approach “the lobster way.” Nobody fully explains why. It just works.

The architecture is a gateway with WebSocket control plane. Skills and plugins let you extend it. The whole thing is on GitHub, MIT licensed.

The Apple Graveyard

Let’s talk about Siri.

“Hey Siri, set a timer for 5 minutes” remains Siri’s peak achievement. Thirteen years after launch, that’s still the most reliable thing it does. Apple has spent the last decade announcing AI features, underdelivering, and hoping nobody notices.

The current state of Apple AI in 2026:

  • $1 billion per year to Google: Apple is literally paying Google for AI models because they couldn’t build competitive ones in-house
  • “Campos” delayed to iOS 27: Their chatbot, announced with great fanfare, won’t ship until at least 2027
  • Unofficial Claude adoption: Bloomberg reports that Anthropic’s Claude is the “unofficial favorite” for future Siri integration

The irony is almost too perfect. Apple’s potential future Siri uses Claude. Clawdbot uses Claude. Clawdbot works today.

Clawdbot showed me what the future of personal AI assistants looks like.

— MacStories feature on Clawdbot

One retired iOS dev shipped what Apple’s billion-dollar AI division couldn’t.

Why This Matters

This isn’t really about Clawdbot vs Siri. It’s about what’s now possible.

  • LLM APIs are commoditized: Claude, GPT, Gemini. They’re all accessible. The hard part isn’t intelligence anymore.

  • Messaging infrastructure exists: WhatsApp, Telegram, iMessage have APIs (official or not). You don’t need to build a new interface.

  • Local execution is viable: A Mac Mini runs this fine. No cloud costs, no data leaving your machine.

  • Open source compounds: 10.8K followers in two months. Community contributions. Shared plugins. The feedback loop is faster than any corporate roadmap.

  • Incumbents aren’t building: Linear and Trello could have shipped an AI-native kanban years ago. They had the data, the users, the infrastructure. They didn’t. Now indie devs are doing it in weekends.

The tinkerer's laboratory

Clawdbot isn’t polished. It’s a power user tool that rewards experimentation. People are discovering capabilities by accident. That’s the energy Apple lost somewhere around iOS 10.

What Clawdbot Doesn’t Solve

The caveats are real:

  • Token costs add up: 180 million tokens isn’t cheap. Heavy users will spend real money on API calls.

  • LLM errors happen: Your bot will occasionally do something dumb. For low-stakes tasks, this is fine. For anything important, verify.

  • Security surface area: Local execution means your machine is the attack surface. Shell access is powerful and dangerous.

  • Not for normies: Setup requires command line comfort. This is a developer tool, not a consumer product.

  • Platform instability: Messaging app APIs change. WhatsApp could break tomorrow. iMessage integration requires workarounds.

The Final Irony

Apple still wins.

People are buying Mac Minis specifically to run the AI assistant Apple couldn’t build. The hardware sales go to Cupertino. Tim Cook doesn’t care why you bought the machine.

But here’s the thing: that’s fine. The point isn’t that Apple loses. The point is that one motivated developer with access to modern LLM APIs can build something useful faster than a trillion-dollar company’s internal AI team.

The moat was always imaginary. Open source just swam across.

Try It

clawd.bot for setup, GitHub for source. You’ll need a Claude API key and comfort with terminal commands.

No tutorial here. The lobster way is figuring it out yourself.

Related Posts